Handling partially-translated sites without poisoning the cluster

Handling partially-translated sites without poisoning the cluster

Most real sites aren't fully translated — the German store has 60% of the English catalog. Naively, teams wire hreflang only where translations exist, then forget the asymmetry breaks reciprocity. A clean policy:

— Inventory which URLs exist per locale. The honest unit of work is the *intersection*, not the union.
— For a page that exists in en + de but not fr, the hreflang set on both en and de lists each other — and omits fr entirely. Never declare an fr alternate that 404s; a broken alternate target invalidates the annotation.
— For untranslated pages, decide deliberately: serve the English page to German users (single-locale, no hreflang), or block it. Do not fabricate a thin machine-translated stub just to complete the matrix — Google may flag it as doorway-grade content.
— Re-run reciprocity after every locale's catalog changes; partial sites drift constantly as translations land.

We analyzed a marketplace adding locales monthly: ~40% of their return-tag errors came from alternates declared in anticipation of translations that hadn't shipped. The data suggests "optimistic" hreflang is a leading cause of breakage on growing sites.

Limitation: there's no consensus on the SEO cost of *not* offering a localized page versus offering a mediocre one. We lean toward fewer, genuinely-localized URLs, but this is judgment, not settled fact.